Out-of-Control Signals, Explained

Out-of-control signals are the patterns on a control chart that are too improbable to blame on routine variation — the chart’s vocabulary for announcing that a special cause has entered the process. A single point beyond the control limits is the most famous signal, but it is only the first of several worth knowing. A process can shift meaningfully and never breach a limit at all; the subtler signals exist to catch exactly that.

How it works

The best-known supplementary rules descend from the Western Electric handbook of 1956, which divided the chart into zones one, two, and three standard deviations from the center line. Beyond the classic point outside the three-sigma limits, the rules flag: two out of three consecutive points beyond two sigma on the same side; four out of five beyond one sigma on the same side; and eight consecutive points on one side of the center line. Other widely used patterns include a long steadily rising or falling trend, cycles that repeat with a rhythm, and points hugging the center line too tightly — which sounds like good news but usually means the subgrouping or the data itself deserves scrutiny.

Every rule added makes the chart more sensitive and raises the rate of false alarms. That trade-off is the reason teams agree on their rule set in advance: a chart whose rules change with the reader’s mood produces arguments, not signals.

A worked example

An electronics assembler charts solder paste height on a production line. After a fixture is swapped during maintenance, no point ever breaches a control limit — but the next nine points all sit below the center line. Under the runs rule, that pattern is far too orderly to be chance, and the chart signals. Measurement confirms the new fixture sits fractionally lower, shifting the whole process down. The fixture is shimmed to spec the same day. A team watching only for points outside the limits would have run for weeks on a subtly shifted process — and found the problem in field returns instead.

• Agree in advance which rules the chart will use, and write them on or beside the chart itself.
• Treat a signal as a question to investigate, never as a verdict on people or a trigger for automatic adjustment.
• Keep an event log with the chart so each signal can be matched against what changed and when.
• Resist running every rule at once on a noisy process — alarm fatigue kills charts faster than any statistical flaw.
• Investigate favorable signals too; a genuine improvement you cannot explain is a gain you cannot keep.

A control chart only works if a signal obliges someone to go and look.

Our Green Belt program ($299, about 35 hours) teaches the primary signals and the discipline of responding to them; Black Belt ($499, about 60 hours) goes deeper into the full rule sets and their false-alarm arithmetic. Both end in proctored exams — 70% to pass, one free retake — where reading a chart correctly is not optional.